ALOHA 开源机械臂（Viper 300& Widow X 250 6DOF机械臂远程操控系统）第四部分

Teleoperation System

远程操作系统

We introduce ALOHA: A Low-cost Open-source Hardware System for Bimanual Teleoperation. With a $20k budget, it is capable of teleoperating precise tasks such as threading a zip tie, dynamic tasks such as juggling a ping pong ball, and contact-rich tasks such as assembling the chain in the NIST board #2.

我们介绍了ALOHA：一种用于双手远程操作的低成本开源硬件系统。凭借2万美元的预算，它能够远程操作精确的任务，如穿拉链，动态任务，如玩乒乓球，以及丰富的接触任务，如在NIST板2中组装链条。

Fig. 1: ALOHA : A Low-cost pen-source Hardware System for Bimanual Teleoperation. The whole system costs $20k with of-the-shelfrobots and 3D printed components Lef: The user teleoperates by backdriving the leader robots. with the follower rohots mirrorine the motionRight: ALOHA is capable of precise. contact-rich, and dynamic tasks. We show examples of both teleoperated and learned skills.

图1:ALOHA：用于双手动远程操作的低成本笔源硬件系统。整个系统的成本约2万美元，包括货架机器人和3D打印组件Lef：用户通过回溯引导机器人进行远程操作。跟随者旋转反光镜的动作右：ALOHA能够精确。接触丰富且动态的任务。我们展示了远程操作和学习技能的例子。

Learned Policy

学习策略

We introduce Action Chunking with Transformers (ACT). The videos below show real-time rollouts of learned policies, imitating from only 50 demonstrations for each task. ACT predicts a sequence of target joint positions given RGB images and proprioception. For the three following tasks, ACT obtains 96%, 84%, 64% success respectively.

我们介绍了变形金刚的动作方块（ACT）。下面的视频显示了学习策略的实时推出，每个任务只模仿了50个演示。ACT预测给定RGB图像和本体感觉的目标关节位置序列。对于以下三项任务，ACT分别获得96%、84%和64%的成功率。

ViperX 6dof Arm (follower)

ViperX 6自由度手臂（从动件）

#Dofs（公司名称）	6+gripper（6+夹具）
Reach（臂展）	750mm
Span（范围）	1500mm
Repeatability（可重复性）	1mm
Accuracy（精确）	5-8mm
Working Payload（工作有效载荷）	750g

Fig. 3: Lef: Camera viewpoints of the front, top, and two wrist cameras, together with an illustration of the bimanual workspace of ALOHAMiddle: Detailed view of the “handle and scissor" mechanism and custom gripers Right: Technical spec of the ViperX 6dof robot 1j.

图3:Lef：正面、顶部和两个手腕摄像头的摄像头视角，以及ALOHAMiddle的双手操作工作区的示意图：“手柄和剪刀”机构和自定义夹具的详细视图右：ViperX 6dof机器人1j的技术规格。

Observations during policy execution

观察报告

We show example image observations (i.e. the input to the ACT policy) at evaluation time. There is a total of 4 RGB cameras each streaming at 480x640. Two of the cameras are stationery and the other two are mounted on the wrist of robots.

我们展示了评估时的示例图像观察结果（即ACT策略的输入）。总共有4个RGB摄像机，每个摄像机的分辨率为480x640。其中两个摄像头是文具，另外两个安装在机器人的手腕上。

Slide Ziploc

幻灯片Ziploc